Re: Is +[NSMutableArray initialize] completely implemented?

[David Ayers]

Andrew Pinski wrote:

> On Feb 22, 2004, at 06:26, David Ayers wrote:
>
> > Kazunobu Kuriyama wrote:
> >
> > > Though this seems to be a dated document, gcc/objc/README of GCC's 
> > > source tells
> > > <quote>
> > > +initialize
> > > ===========
> > > This method, if defined, is called before any other instance or class 
> > > methods of
> > > that particular class.  This method is not inherited, and is thus not 
> > > called as
> > > initializer for a subclass that doesn't define it itself.  Thus, each 
> > > +initialize
> > > method is called once (or never if no methods of that particular 
> > > class is never
> > > called).
> > > Besides this, it is allowed ot have several +initialize methods, one 
> > > for each
> > > category.  The order in which these (multiple methods) are called is 
> > > not well
> > > defined.  I am not completely certain what the semantics of this 
> > > methods is
> > > for other implementations, but this is how it works for GNU Objective C.
> > > <unquote>
> >
> >
> > Ahh!  Thanks, that's probably where I had read it and was later 
> > confused to find that it doesn't work that way.  AFAIK the Apple/NeXT 
> > runtimes also don't call +initialize per category.  So I think the 
> > comment needs to be fixed.
[snip]
> Can you file a bug <http://gcc.gnu.org/bugs.html>, I think the intended 
> behavior
> is to have them both called.  I will be looking into this.

Well, I can, and if only for documentation purposes.  But I have been 
able to verify that the NeXT/Apple Runtime behave the same way.

Personally I would like the feature that +initialize be called on a per 
category basis, as it makes some things a whole lot easier for me (esp. 
wrt to GDL2, (if we could change this for gcc 3.5 and make it a 
prerequisite for GDL2... haha!))

But it does have implications wrt being able to overwrite the default 
implementation.  Yet I agree, that that is rather dubious and will 
hardly ever be able to do the Right Thing, if you consider that usually 
you just insure that related classes get initialized and static locals 
get initialized that you can't set from outside the compilation unit anyway.

Yet I think this is an issue for objc-language@ so I'll prepare a post. 
 I strongly think it would only make sense if we change it in both 
runtimes if we change it at all.

Cheers,
David

PS: Didn't get around to the bug report and formulating the 
objc-language@ post today, but I should on the weekend.